Maritime domain awareness: Engineering analytic frameworks

Maritime Domain Awareness is critical for protecting sea lanes, ports, harbors, offshore structures like oil and gas rigs and other types of critical infrastructure against common threats and illegal activities. Limited surveillance resources constrain maritime domain awareness and compromise full safety and security coverage at all times. This situation calls for innovative intelligent systems for interactive situation analysis and decision support to assist marine authorities in their routine surveillance operations. The main contribution of this thesis is the formal engineering of robust methodical frameworks for the development of intelligent algorithms, systems, and services for real-time situation analysis, anomaly detection, and decision support; particularly for the maritime domain. Best engineering practice calls for a system to be modeled prior to construction, so one can rigorously inspect and reason about the key system properties, making sure these are both well understood and properly established. We therefore use Abstract State Machine as a Formal Method for engineering these frameworks to address the main aspects of maritime domain awareness including situation analysis and anomaly detection as well as decision support and emergency response. To be more specific, we propose a novel situation analysis approach to analyze marine traffic data and differentiate various scenarios of vessel engagement for the purpose of detecting anomalies of interest for marine vessels that operate over some period of time in relative proximity to each other. We consider such scenarios as probabilistic processes and analyze complex vessel trajectories using machine learning to model common patterns. Specifically, we represent patterns as left-to-right Hidden Markov Models and classify them using Support Vector Machines. To differentiate suspicious activities from unobjectionable behavior, we explore fusion of data and information, including kinematic features, geospatial features, contextual information and maritime domain knowledge. Our experimental evaluation shows the effectiveness of the proposed approach using comprehensive real-world vessel tracking data from coastal waters of North America.